also rename CRPS to CRPS_FROM_QS, and modify the algorithm for outputs
to CSV to deal with multisite case required to support metric ES

the outputting functionality was not looping through the diagnostics

giving access to "completeness" information

needed for REL_DIAG

`evalhyd` recently inherited the mask generating functionality from
user-provided masking conditions, so the internal functionality is
removed. This functionality was only accessible to `evalp`, this is
now accessible to both `evalp` and `evald`.

The algorithm was overwriting the same CSV file for each temporal subset
in the output array, which meant that only the last subset was available
in files. This is now fixed, but in keeping all subsets in the CSV
files, some output arrays are 3D (that is the case for multi-component
metrics such as BS_CRD or BS_LBD). Since CSV files are intrinsically 2D
(i.e. lines and columns only), subsets are stacked on top of one another
in the CSV output files, so that it remains 2D.

since the dimensions are inferred arbitrarily from the first
prediction file, later files' data may have been larger in size, and
silently truncated, so a check was required to may sure data was not
ignored silently

which includes the addition of a "sites" axis in the dimensions of
streamflow thresholds and temporal subsets (i.e. masks)

New parameter "gen_mask" expects a string specifying conditions to
generate masks used to perform the temporal subsets. Conditions can be
based on streamflow (e.g. q{>9} {q<=10} q{<9,>=10} etc.) or on time
indices (e.g. t{4,5,6} t{10:30} t(11,12,13,20:50,88} etc.).

For now, the user is expected to produce a CSV file of ones/zeros
to specify the masks to use for temporal subsetting.

which includes the addition of sites and leadtimes dimensions for
evalp, the addition of the metrics RMSE/KGE/KGEPRIME for evald, and
the addition of the transformation functionality for evald.